Diabetic Retinopathy (DR) remains a leading global cause of vision impairment, and early and precise identification of retinal vascular structures plays a pivotal role in diagnosis and disease progression monitoring. However, retinal vessel segmentation from fundus images remains challenging because of complex vascular morphology, large variations in vessel width, pathological lesions and artifacts, and low vessel-to-background contrast. While deep learning has shown promise, existing models still struggle with fine-grained vessel segmentation, particularly in the presence of lesions and imaging artifacts. To address these challenges, RVSM (Retinal Vessel Segmentation Model) is introduced as a deep learning architecture specifically designed for retinal fundus image analysis. The proposed framework integrates a Convolutional Block Attention Module (CBAM) to enhance both local feature representation and global contextual modeling, enabling improved delineation of ambiguous vessel boundaries and better capture of long-range dependencies. Through attention-driven feature refinement, the architecture emphasizes clinically relevant spatial regions and supports robust segmentation under challenging pathological conditions. The effectiveness of the proposed approach is evaluated on three benchmark datasets, namely DRIVE, STARE, and CHASEDB1. The model achieves segmentation accuracies of 96.32%, 94.43%, and 93.22%, respectively, while also demonstrating strong Dice and IoU performance. In addition, the framework supports downstream diabetic retinopathy classification, achieving accuracies of 97.31%, 95.45%, and 94.46% across the same datasets. These results indicate that integrating attention-driven mechanisms into task-specific architectures can improve both segmentation robustness and diagnostic effectiveness. Overall, RVSM provides a scalable and clinically relevant solution for retinal vessel analysis in diabetic retinopathy screening and follow-up.
Rehman et al. (Sat,) studied this question.